“A significant portion of our health remains imperceptible even to us, and a primary obstacle to improving health is merely being unaware of what’s happening,” explains Apple’s Steve Waydo, director for health sensing.
Given that February marks Heart Month, it’s fitting to discuss the team responsible for creating the new hypertension notification system for watchOS 26 and the Apple Watch.
My conversation with Steve Waydo, Apple’s director for health sensing, and Dr. Rajiv Kumar, a physician-researcher, provided insight into the scientific foundation and strategic choices guiding their extensive initiative to equip smartwatch users with a dependable and practical tool for monitoring heart health.
Waydo spearheaded the prolonged development of the hypertension notification functionality. He noted, “The concept emerged not long after the debut of the initial Apple Watch. We possessed a device continuously gathering physiological data from users, an unprecedented capability. We recognized this as a novel and distinct opportunity.”
However, a prerequisite for Apple was to innovate new sensor technologies, gather top-tier technical and medical experts, and construct precise and efficient machine learning systems. Driven by a desire for scientifically robust solutions, Apple also initiated a comprehensive heart health study in collaboration with the University of Michigan.
Demystifying Hypertension
Hypertension signifies a persistent state of elevated blood pressure. With every heartbeat, blood is propelled from the heart into the arteries. When blood pressure is high, significant resistance causes the heart to exert more effort than usual to circulate blood. This sustained exertion eventually leads to hypertension. The challenge with this ailment is its complete lack of symptoms, earning it the moniker of “silent killer.”
Over one billion individuals worldwide are affected by this condition, with almost half of all adults in the US experiencing it. Astonishingly, roughly half of these affected individuals remain unaware of their diagnosis.
This underscores the significance of Apple’s innovative tool, which holds the potential to assist individuals in recognizing the condition and adopting proactive management strategies. As Waydo remarked, “Much of our physiological state is hidden from our own perception, and a primary impediment to improved well-being is simply a lack of awareness.”
Apple aims to leverage the capabilities of wearable technology to uncover health conditions like hypertension, which might otherwise prove challenging for individuals to manage effectively on their own.
The Synergy of Machine Learning, Data, and Context
The data generated by an Apple Watch stands apart from typical diagnostic data due to its continuous, near-daily wear. This constant monitoring allows the collected information to evolve, facilitating the discovery of profound health insights. Artificial intelligence further enhances the actionability of this data, empowering the device to autonomously highlight valuable observations derived from its tracked metrics.
Kumar elaborated on Apple’s creation of a machine learning system designed to integrate personal health data with real-world insights gathered from the Apple Heart study. This study proved instrumental in enabling Apple to comprehend “the nature of these signals, how they evolve throughout an individual’s life, and across diverse situations, ultimately dissecting raw sensor data into thousands of measurable, independent factors.”
Furthermore, Apple utilized supervised learning data, specifically information sourced from both sensor readings and validated ground truth. Such data is a product of Apple’s collaboration with the University of Michigan. The significant advantage of this fusion lies in Apple’s ability to observe the correlation between sensor data and scientific findings. Consequently, machine learning models can process personal data and compare it against sensor data across numerous variables to ascertain an individual’s hypertensive state.
For a deeper understanding of Apple’s system, readers can consult the company’s comprehensive white paper dedicated to this subject.
The Apple Watch: Your Wearable Health Companion
Kumar stated, “These machine learning algorithms are a crucial facilitating technology, as conditions like hypertension present themselves through incredibly nuanced patterns in our physiological signals. We’re observing minute characteristics within the actual waveform from the sensors… Our focus is on detecting far more subtle indicators linked to high blood pressure, as these signals reveal insights into your blood vessels’ responses with each heartbeat. Consequently, we deploy these machine learning methodologies across millions of data points.”
Waydo remarked, “Having been with Apple for 13 years, I’ve witnessed this entire progression. These very same tools empower your watch to monitor your activity, discern whether you’re walking or swimming, approximate your duration in various sleep stages, and detect falls, enabling it to contact emergency services. Indeed, we integrate machine learning extensively.”
Apple consistently observes that it is essential to analyze how an individual’s data changes over an extended duration, rather than issuing alerts based on isolated instances.
Navigating Data Noise
While the adage “garbage in, garbage out” frequently applies in the era of AI, Apple’s specialists offered intriguing perspectives on the characteristics of data noise. Kumar noted, “We process immense volumes of data to develop these functionalities.” This necessitates that the algorithms accurately assess the quality of the data they acquire.
Apple, by augmenting its data with findings from extensive, real-world studies, discovered that embracing “messiness” can lead to superior outcomes. He explained, “Possessing a dataset that genuinely embodies complexity and authenticity is crucial for generating signals that transcend mere research interest and can genuinely be applied to real individuals utilizing our devices globally.”
AI Development: Iteration as a Path to Success
Achieving system functionality involved a protracted cycle of iteration and refinement. Apple’s teams continuously expanded and enhanced their datasets, refined algorithms, and persistently improved their creations until they were deemed suitable for public release. He described the process: “We repeatedly perform that cycle for weeks, months, or even years until we achieve a satisfactory outcome.”
Interestingly, the team also finds enthusiasm in system failures. “We might encounter specific use cases, user types, or scenarios resulting in numerous false positives or a complete absence of true positives. This guides our efforts to refine and iterate on the algorithm, typically by acquiring additional data that encompasses these particular use cases for our machine learning training.”
In their development process, Apple’s teams meticulously examined demographic factors. The objective was to guarantee that the efficacy of Apple’s systems remains unaffected by age, sex, or race. As a global corporation serving a vast user base, Apple’s solutions must be universally functional.
Important Caveats
This feature is not designed to fully supersede routine medical examinations. Recent reports, such as the one stating not all instances of hypertension will be detected, are accurate and highlight the delicate balance developers sought to achieve in deploying the system. This stems from the team’s understanding that making the algorithm more sensitive would identify more cases, but at the expense of an increased number of false positives.
The risk associated with false positives is that users may begin to disregard alerts. Fundamentally, health notifications from a device require unwavering trust in their precision. No individual desires to receive erroneous data.
Kumar elucidated the dilemma: whether Apple should pursue absolute sensitivity, thereby accepting a high incidence of false positives, or develop a system that significantly reduces them. This necessitated Apple striking a careful balance.
“Our efforts are heavily biased towards mitigating false positives and ensuring that any notification of a potential issue—be it hypertension, an irregular rhythm, or other concerns—is genuinely reliable,” he stated. “Occasionally, this implies that certain cases might not be detected, because attempting to capture these additional cases would also lead to notifying many individuals who don’t require it, thereby diminishing the overall effectiveness of the system.”
Apple advises any Apple Watch user who receives a hypertension notification to monitor and record their blood pressure, and to consult their physician.
User Testimonials
Due to Apple’s unwavering dedication to privacy, its development teams are unable to monitor the real-world effectiveness of its systems, as they do not access personal device data. Nevertheless, the team frequently receives correspondence from users, healthcare professionals, and other impacted individuals.
Waydo expressed, “It’s immensely gratifying to hear from clinicians who recount meeting patients who would otherwise have remained undiagnosed or not sought care, and how our technology genuinely transformed their lives. Every feature we develop, be it for women’s health, hearing health, or cardiac health, is fundamentally rooted in science, designed to be actionable, and unequivocally built with privacy as its paramount principle.”
Consequently, as individuals acquire a deeper understanding of their personal health, they are empowered to make more informed and beneficial health choices going forward.
Connect with me on Twitter, or become a member of the AppleHolic’s bar & grill and Apple Discussions communities on MeWe. You can also find me on Mastodon.
